CROSCIM dataset: 15-day extract (2022-07-08 to 2022-07-22) and preprocessed training samples

Extract of the CROSCIM dataset: Arctic sea-ice observations, simulated satellite products (CIMR, CRISTAL), model outputs and atmospheric forcing on common polar stereographic grids. Daily gridded files (2022-07-08 to 2022-07-22, 15 days), one .tar archive per folder, NetCDF-4: ASIP_L3.tar: DMI ASIP Arctic L3 sea-ice mosaics (sea-ice concentration, stage of development, floe size), EPSG:3411. atm_data.tar: ECMWF ERA5 atmospheric fields regridded to the 5 km grid (t2m, d2m, msl, u10, v10, tcc, ssrd, strd, tp). data_noise.tar: simulated noisy CIMR retrievals (SIC, SIT, Tsurf). out_CRISTAL.tar: simulated CRISTAL products (SIT, snow depth, SSH) with the corresponding model fields. out_MOD.tar: model output (SIC, SIT, snow depth, SSH). Preprocessed deep-learning samples (2021): preproc_CROSCIM_x2.nc, preproc_CROSCIM_x10.nc, preproc_CROSCIM_x50.nc, at 1 km, 5 km and 25 km resolution respectively (fields down-pooled by a factor 2, 10 and 50). Each sample is a 256 x 256 patch over 15 time steps, with inputs (ASIP, CIMR, CRISTAL, model, atmosphere) and targets (tgt_SIC, tgt_SIT). The original files hold 3088 samples; 320 samples per file were kept, evenly spaced by start date over 2021, with the same indices in the three files (listed in preproc_selected_samples.txt). Values, compression and chunking are unchanged. See README.md for the full description of every file and variable, and visualize_preproc.py for an example of loading the samples. Please also cite the original products: DMI ASIP sea-ice products (Danish Meteorological Institute) and ERA5 (Hersbach et al., 2020, doi:10.1002/qj.3803; contains modified Copernicus Climate Change Service information).